Skill UI

A comprehensive guide to understanding the complete state (context) available to Large Language Models during inference. Covers core components like System Prompts, Tool Definitions, Message History, and Retrieval Augmented Generation (RAG). Learning to engineer context is crucial for building robust, efficient, and reliable AI agent systems by managing the attention budget and ensuring high-signal information delivery.

Orchestrates PyTorch-native TorchTitan 4D parallelism (FSDP2, TP, PP, CP) for large-scale LLM pretraining, scaling from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing to boot Llama 3.1, DeepSeek V3, or custom models.

Evaluates large language models across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag) with standardized prompts and metrics, making it easy to benchmark HuggingFace or API models, compare releases, and track training progress.

This guide details the use of the GGUF format and quantization techniques for running large language models (LLMs) efficiently. GGUF enables optimal inference on consumer hardware, Apple Silicon, and systems requiring CPU-only operation. By compressing models using various K-quant methods (like Q4_K_M), developers can significantly reduce memory footprint and hardware requirements while maintaining high performance.

This skill provides comprehensive methods to significantly reduce expenditure when using the Ideogram API for image generation. Techniques include implementing two-phase generation workflows (drafting with cheap models, finalizing with premium models), batching multiple images in a single API call, caching identical prompts, and setting up robust budget tracking alerts. Ideal for developers managing large-scale creative AI projects.

LitGPT provides clean, readable, and production-ready implementations for over 20 state-of-the-art LLMs (including Llama, Gemma, Phi, Mistral). It supports comprehensive workflows, including full fine-tuning, efficient LoRA/QLoRA adaptation, and pretraining from scratch. Ideal for developers needing academic understanding or robust, deployable model architectures.

Knowledge Distillation is a crucial technique used to compress large, high-performing Language Models (Teacher) into smaller, more efficient Student models. This allows developers to deploy powerful AI capabilities (like those of GPT-4) on resource-constrained devices while significantly reducing inference costs. The method uses advanced techniques such as soft targets, temperature scaling, and Reverse KL Divergence (MiniLLM) to transfer knowledge effectively.

Integrate Langfuse into CI/CD pipelines (e.g., GitHub Actions) to automate the testing and validation of Large Language Models (LLMs). This guide covers prompt regression testing, tracing validation, and setting up experiment-driven quality gates, ensuring reliable and high-performing AI applications throughout the development lifecycle.

Learn how to monitor, analyze, and optimize the expenditure of Large Language Models (LLMs) using Langfuse's advanced analytics suite. This guide covers techniques like automatic token tracking, programmatic cost reporting via the Metrics API, and implementing smart model routing to automatically select the most cost-effective model based on task complexity and input length. Essential for any developer managing AI budgets and ensuring operational efficiency.

This skill simulates an expert prompt engineer, specializing in optimizing prompts for Large Language Models (LLMs). It applies advanced techniques such as Constitutional AI, Chain-of-Thought reasoning, and model-specific tuning. Use this tool to transform basic instructions into production-ready, high-performance prompts, significantly boosting accuracy, reducing hallucinations, and optimizing token usage.

Compress large language models quickly using Wanda, SparseGPT, and N:M pruning to enforce hardware-friendly sparsity, enabling faster inference and deployment without retraining.

This skill provides comprehensive guidance on training Mixture of Experts (MoE) models, including architectures like Mixtral and DeepSeek-V3. It details techniques for achieving large-scale model capacity with significantly reduced computational costs (e.g., 5x reduction vs. dense models). Topics covered include top-k routing, load balancing, expert parallelism, and integration with frameworks like DeepSpeed and HuggingFace for efficient, resource-constrained training.